2 Copyright (C) 1996-2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
55 # include "http-ntlm.h"
63 extern char *version_string;
66 # define MIN(x, y) ((x) > (y) ? (y) : (x))
70 static bool cookies_loaded_p;
71 static struct cookie_jar *wget_cookie_jar;
73 #define TEXTHTML_S "text/html"
74 #define TEXTXHTML_S "application/xhtml+xml"
76 /* Some status code validation macros: */
77 #define H_20X(x) (((x) >= 200) && ((x) < 300))
78 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
79 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
80 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
81 || (x) == HTTP_STATUS_SEE_OTHER \
82 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
84 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
86 #define HTTP_STATUS_OK 200
87 #define HTTP_STATUS_CREATED 201
88 #define HTTP_STATUS_ACCEPTED 202
89 #define HTTP_STATUS_NO_CONTENT 204
90 #define HTTP_STATUS_PARTIAL_CONTENTS 206
92 /* Redirection 3xx. */
93 #define HTTP_STATUS_MULTIPLE_CHOICES 300
94 #define HTTP_STATUS_MOVED_PERMANENTLY 301
95 #define HTTP_STATUS_MOVED_TEMPORARILY 302
96 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
97 #define HTTP_STATUS_NOT_MODIFIED 304
98 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
100 /* Client error 4xx. */
101 #define HTTP_STATUS_BAD_REQUEST 400
102 #define HTTP_STATUS_UNAUTHORIZED 401
103 #define HTTP_STATUS_FORBIDDEN 403
104 #define HTTP_STATUS_NOT_FOUND 404
105 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
107 /* Server errors 5xx. */
108 #define HTTP_STATUS_INTERNAL 500
109 #define HTTP_STATUS_NOT_IMPLEMENTED 501
110 #define HTTP_STATUS_BAD_GATEWAY 502
111 #define HTTP_STATUS_UNAVAILABLE 503
114 rel_none, rel_name, rel_value, rel_both
121 struct request_header {
123 enum rp release_policy;
125 int hcount, hcapacity;
128 /* Create a new, empty request. At least request_set_method must be
129 called before the request can be used. */
131 static struct request *
134 struct request *req = xnew0 (struct request);
136 req->headers = xnew_array (struct request_header, req->hcapacity);
140 /* Set the request's method and its arguments. METH should be a
141 literal string (or it should outlive the request) because it will
142 not be freed. ARG will be freed by request_free. */
145 request_set_method (struct request *req, const char *meth, char *arg)
151 /* Return the method string passed with the last call to
152 request_set_method. */
155 request_method (const struct request *req)
160 /* Free one header according to the release policy specified with
161 request_set_header. */
164 release_header (struct request_header *hdr)
166 switch (hdr->release_policy)
183 /* Set the request named NAME to VALUE. Specifically, this means that
184 a "NAME: VALUE\r\n" header line will be used in the request. If a
185 header with the same name previously existed in the request, its
186 value will be replaced by this one. A NULL value means do nothing.
188 RELEASE_POLICY determines whether NAME and VALUE should be released
189 (freed) with request_free. Allowed values are:
191 - rel_none - don't free NAME or VALUE
192 - rel_name - free NAME when done
193 - rel_value - free VALUE when done
194 - rel_both - free both NAME and VALUE when done
196 Setting release policy is useful when arguments come from different
197 sources. For example:
199 // Don't free literal strings!
200 request_set_header (req, "Pragma", "no-cache", rel_none);
202 // Don't free a global variable, we'll need it later.
203 request_set_header (req, "Referer", opt.referer, rel_none);
205 // Value freshly allocated, free it when done.
206 request_set_header (req, "Range",
207 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
212 request_set_header (struct request *req, char *name, char *value,
213 enum rp release_policy)
215 struct request_header *hdr;
220 /* A NULL value is a no-op; if freeing the name is requested,
221 free it now to avoid leaks. */
222 if (release_policy == rel_name || release_policy == rel_both)
227 for (i = 0; i < req->hcount; i++)
229 hdr = &req->headers[i];
230 if (0 == strcasecmp (name, hdr->name))
232 /* Replace existing header. */
233 release_header (hdr);
236 hdr->release_policy = release_policy;
241 /* Install new header. */
243 if (req->hcount >= req->hcapacity)
245 req->hcapacity <<= 1;
246 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
248 hdr = &req->headers[req->hcount++];
251 hdr->release_policy = release_policy;
254 /* Like request_set_header, but sets the whole header line, as
255 provided by the user using the `--header' option. For example,
256 request_set_user_header (req, "Foo: bar") works just like
257 request_set_header (req, "Foo", "bar"). */
260 request_set_user_header (struct request *req, const char *header)
263 const char *p = strchr (header, ':');
266 BOUNDED_TO_ALLOCA (header, p, name);
270 request_set_header (req, xstrdup (name), (char *) p, rel_name);
273 /* Remove the header with specified name from REQ. Returns true if
274 the header was actually removed, false otherwise. */
277 request_remove_header (struct request *req, char *name)
280 for (i = 0; i < req->hcount; i++)
282 struct request_header *hdr = &req->headers[i];
283 if (0 == strcasecmp (name, hdr->name))
285 release_header (hdr);
286 /* Move the remaining headers by one. */
287 if (i < req->hcount - 1)
288 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
296 #define APPEND(p, str) do { \
297 int A_len = strlen (str); \
298 memcpy (p, str, A_len); \
302 /* Construct the request and write it to FD using fd_write. */
305 request_send (const struct request *req, int fd)
307 char *request_string, *p;
308 int i, size, write_error;
310 /* Count the request size. */
313 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
314 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
316 for (i = 0; i < req->hcount; i++)
318 struct request_header *hdr = &req->headers[i];
319 /* NAME ": " VALUE "\r\n" */
320 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
326 p = request_string = alloca_array (char, size);
328 /* Generate the request. */
330 APPEND (p, req->method); *p++ = ' ';
331 APPEND (p, req->arg); *p++ = ' ';
332 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
334 for (i = 0; i < req->hcount; i++)
336 struct request_header *hdr = &req->headers[i];
337 APPEND (p, hdr->name);
338 *p++ = ':', *p++ = ' ';
339 APPEND (p, hdr->value);
340 *p++ = '\r', *p++ = '\n';
343 *p++ = '\r', *p++ = '\n', *p++ = '\0';
344 assert (p - request_string == size);
348 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
350 /* Send the request to the server. */
352 write_error = fd_write (fd, request_string, size - 1, -1);
354 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
359 /* Release the resources used by REQ. */
362 request_free (struct request *req)
365 xfree_null (req->arg);
366 for (i = 0; i < req->hcount; i++)
367 release_header (&req->headers[i]);
368 xfree_null (req->headers);
372 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
373 PROMISED_SIZE bytes are sent over the wire -- if the file is
374 longer, read only that much; if the file is shorter, report an error. */
377 post_file (int sock, const char *file_name, wgint promised_size)
379 static char chunk[8192];
384 DEBUGP (("[writing POST file %s ... ", file_name));
386 fp = fopen (file_name, "rb");
389 while (!feof (fp) && written < promised_size)
392 int length = fread (chunk, 1, sizeof (chunk), fp);
395 towrite = MIN (promised_size - written, length);
396 write_error = fd_write (sock, chunk, towrite, -1);
406 /* If we've written less than was promised, report a (probably
407 nonsensical) error rather than break the promise. */
408 if (written < promised_size)
414 assert (written == promised_size);
415 DEBUGP (("done]\n"));
419 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
420 If so, return the pointer to the position after the line, otherwise
421 return NULL. This is used as callback to fd_read_hunk. The data
422 between START and PEEKED has been read and cannot be "unread"; the
423 data after PEEKED has only been peeked. */
426 response_head_terminator (const char *start, const char *peeked, int peeklen)
430 /* If at first peek, verify whether HUNK starts with "HTTP". If
431 not, this is a HTTP/0.9 request and we must bail out without
433 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
436 /* Look for "\n[\r]\n", and return the following position if found.
437 Start two chars before the current to cover the possibility that
438 part of the terminator (e.g. "\n\r") arrived in the previous
440 p = peeked - start < 2 ? start : peeked - 2;
441 end = peeked + peeklen;
443 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
444 for (; p < end - 2; p++)
447 if (p[1] == '\r' && p[2] == '\n')
449 else if (p[1] == '\n')
452 /* p==end-2: check for \n\n directly preceding END. */
453 if (p[0] == '\n' && p[1] == '\n')
459 /* The maximum size of a single HTTP response we care to read. Rather
460 than being a limit of the reader implementation, this limit
461 prevents Wget from slurping all available memory upon encountering
462 malicious or buggy server output, thus protecting the user. Define
463 it to 0 to remove the limit. */
465 #define HTTP_RESPONSE_MAX_SIZE 65536
467 /* Read the HTTP request head from FD and return it. The error
468 conditions are the same as with fd_read_hunk.
470 To support HTTP/0.9 responses, this function tries to make sure
471 that the data begins with "HTTP". If this is not the case, no data
472 is read and an empty request is returned, so that the remaining
473 data can be treated as body. */
476 read_http_response_head (int fd)
478 return fd_read_hunk (fd, response_head_terminator, 512,
479 HTTP_RESPONSE_MAX_SIZE);
483 /* The response data. */
486 /* The array of pointers that indicate where each header starts.
487 For example, given this HTTP response:
494 The headers are located like this:
496 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
498 headers[0] headers[1] headers[2] headers[3]
500 I.e. headers[0] points to the beginning of the request,
501 headers[1] points to the end of the first header and the
502 beginning of the second one, etc. */
504 const char **headers;
507 /* Create a new response object from the text of the HTTP response,
508 available in HEAD. That text is automatically split into
509 constituent header lines for fast retrieval using
512 static struct response *
513 resp_new (const char *head)
518 struct response *resp = xnew0 (struct response);
523 /* Empty head means that we're dealing with a headerless
524 (HTTP/0.9) response. In that case, don't set HEADERS at
529 /* Split HEAD into header lines, so that resp_header_* functions
530 don't need to do this over and over again. */
536 DO_REALLOC (resp->headers, size, count + 1, const char *);
537 resp->headers[count++] = hdr;
539 /* Break upon encountering an empty line. */
540 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
543 /* Find the end of HDR, including continuations. */
546 const char *end = strchr (hdr, '\n');
552 while (*hdr == ' ' || *hdr == '\t');
554 DO_REALLOC (resp->headers, size, count + 1, const char *);
555 resp->headers[count] = NULL;
560 /* Locate the header named NAME in the request data, starting with
561 position START. This allows the code to loop through the request
562 data, filtering for all requests of a given name. Returns the
563 found position, or -1 for failure. The code that uses this
564 function typically looks like this:
566 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
567 ... do something with header ...
569 If you only care about one header, use resp_header_get instead of
573 resp_header_locate (const struct response *resp, const char *name, int start,
574 const char **begptr, const char **endptr)
577 const char **headers = resp->headers;
580 if (!headers || !headers[1])
583 name_len = strlen (name);
589 for (; headers[i + 1]; i++)
591 const char *b = headers[i];
592 const char *e = headers[i + 1];
594 && b[name_len] == ':'
595 && 0 == strncasecmp (b, name, name_len))
598 while (b < e && ISSPACE (*b))
600 while (b < e && ISSPACE (e[-1]))
610 /* Find and retrieve the header named NAME in the request data. If
611 found, set *BEGPTR to its starting, and *ENDPTR to its ending
612 position, and return true. Otherwise return false.
614 This function is used as a building block for resp_header_copy
615 and resp_header_strdup. */
618 resp_header_get (const struct response *resp, const char *name,
619 const char **begptr, const char **endptr)
621 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
625 /* Copy the response header named NAME to buffer BUF, no longer than
626 BUFSIZE (BUFSIZE includes the terminating 0). If the header
627 exists, true is returned, false otherwise. If there should be no
628 limit on the size of the header, use resp_header_strdup instead.
630 If BUFSIZE is 0, no data is copied, but the boolean indication of
631 whether the header is present is still returned. */
634 resp_header_copy (const struct response *resp, const char *name,
635 char *buf, int bufsize)
638 if (!resp_header_get (resp, name, &b, &e))
642 int len = MIN (e - b, bufsize - 1);
643 memcpy (buf, b, len);
649 /* Return the value of header named NAME in RESP, allocated with
650 malloc. If such a header does not exist in RESP, return NULL. */
653 resp_header_strdup (const struct response *resp, const char *name)
656 if (!resp_header_get (resp, name, &b, &e))
658 return strdupdelim (b, e);
661 /* Parse the HTTP status line, which is of format:
663 HTTP-Version SP Status-Code SP Reason-Phrase
665 The function returns the status-code, or -1 if the status line
666 appears malformed. The pointer to "reason-phrase" message is
667 returned in *MESSAGE. */
670 resp_status (const struct response *resp, char **message)
677 /* For a HTTP/0.9 response, assume status 200. */
679 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
683 p = resp->headers[0];
684 end = resp->headers[1];
690 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
694 /* Match the HTTP version. This is optional because Gnutella
695 servers have been reported to not specify HTTP version. */
696 if (p < end && *p == '/')
699 while (p < end && ISDIGIT (*p))
701 if (p < end && *p == '.')
703 while (p < end && ISDIGIT (*p))
707 while (p < end && ISSPACE (*p))
709 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
712 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
717 while (p < end && ISSPACE (*p))
719 while (p < end && ISSPACE (end[-1]))
721 *message = strdupdelim (p, end);
727 /* Release the resources used by RESP. */
730 resp_free (struct response *resp)
732 xfree_null (resp->headers);
736 /* Print the server response, line by line, omitting the trailing CRLF
737 from individual header lines, and prefixed with PREFIX. */
740 print_server_response (const struct response *resp, const char *prefix)
745 for (i = 0; resp->headers[i + 1]; i++)
747 const char *b = resp->headers[i];
748 const char *e = resp->headers[i + 1];
750 if (b < e && e[-1] == '\n')
752 if (b < e && e[-1] == '\r')
754 /* This is safe even on printfs with broken handling of "%.<n>s"
755 because resp->headers ends with \0. */
756 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
760 /* Parse the `Content-Range' header and extract the information it
761 contains. Returns true if successful, false otherwise. */
763 parse_content_range (const char *hdr, wgint *first_byte_ptr,
764 wgint *last_byte_ptr, wgint *entity_length_ptr)
768 /* Ancient versions of Netscape proxy server, presumably predating
769 rfc2068, sent out `Content-Range' without the "bytes"
771 if (0 == strncasecmp (hdr, "bytes", 5))
774 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
778 while (ISSPACE (*hdr))
785 for (num = 0; ISDIGIT (*hdr); hdr++)
786 num = 10 * num + (*hdr - '0');
787 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
789 *first_byte_ptr = num;
791 for (num = 0; ISDIGIT (*hdr); hdr++)
792 num = 10 * num + (*hdr - '0');
793 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
795 *last_byte_ptr = num;
797 for (num = 0; ISDIGIT (*hdr); hdr++)
798 num = 10 * num + (*hdr - '0');
799 *entity_length_ptr = num;
803 /* Read the body of the request, but don't store it anywhere and don't
804 display a progress gauge. This is useful for reading the bodies of
805 administrative responses to which we will soon issue another
806 request. The response is not useful to the user, but reading it
807 allows us to continue using the same connection to the server.
809 If reading fails, false is returned, true otherwise. In debug
810 mode, the body is displayed for debugging purposes. */
813 skip_short_body (int fd, wgint contlen)
816 SKIP_SIZE = 512, /* size of the download buffer */
817 SKIP_THRESHOLD = 4096 /* the largest size we read */
819 char dlbuf[SKIP_SIZE + 1];
820 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
822 /* We shouldn't get here with unknown contlen. (This will change
823 with HTTP/1.1, which supports "chunked" transfer.) */
824 assert (contlen != -1);
826 /* If the body is too large, it makes more sense to simply close the
827 connection than to try to read the body. */
828 if (contlen > SKIP_THRESHOLD)
831 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
835 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
838 /* Don't normally report the error since this is an
839 optimization that should be invisible to the user. */
840 DEBUGP (("] aborting (%s).\n",
841 ret < 0 ? fd_errstr (fd) : "EOF received"));
845 /* Safe even if %.*s bogusly expects terminating \0 because
846 we've zero-terminated dlbuf above. */
847 DEBUGP (("%.*s", ret, dlbuf));
850 DEBUGP (("] done.\n"));
854 /* Persistent connections. Currently, we cache the most recently used
855 connection as persistent, provided that the HTTP server agrees to
856 make it such. The persistence data is stored in the variables
857 below. Ideally, it should be possible to cache an arbitrary fixed
858 number of these connections. */
860 /* Whether a persistent connection is active. */
861 static bool pconn_active;
864 /* The socket of the connection. */
867 /* Host and port of the currently active persistent connection. */
871 /* Whether a ssl handshake has occoured on this connection. */
874 /* Whether the connection was authorized. This is only done by
875 NTLM, which authorizes *connections* rather than individual
876 requests. (That practice is peculiar for HTTP, but it is a
877 useful optimization.) */
881 /* NTLM data of the current connection. */
882 struct ntlmdata ntlm;
886 /* Mark the persistent connection as invalid and free the resources it
887 uses. This is used by the CLOSE_* macros after they forcefully
888 close a registered persistent connection. */
891 invalidate_persistent (void)
893 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
894 pconn_active = false;
895 fd_close (pconn.socket);
900 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
901 persistent. This will enable someone to use the same connection
902 later. In the context of HTTP, this must be called only AFTER the
903 response has been received and the server has promised that the
904 connection will remain alive.
906 If a previous connection was persistent, it is closed. */
909 register_persistent (const char *host, int port, int fd, bool ssl)
913 if (pconn.socket == fd)
915 /* The connection FD is already registered. */
920 /* The old persistent connection is still active; close it
921 first. This situation arises whenever a persistent
922 connection exists, but we then connect to a different
923 host, and try to register a persistent connection to that
925 invalidate_persistent ();
931 pconn.host = xstrdup (host);
934 pconn.authorized = false;
936 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
939 /* Return true if a persistent connection is available for connecting
943 persistent_available_p (const char *host, int port, bool ssl,
944 bool *host_lookup_failed)
946 /* First, check whether a persistent connection is active at all. */
950 /* If we want SSL and the last connection wasn't or vice versa,
951 don't use it. Checking for host and port is not enough because
952 HTTP and HTTPS can apparently coexist on the same port. */
953 if (ssl != pconn.ssl)
956 /* If we're not connecting to the same port, we're not interested. */
957 if (port != pconn.port)
960 /* If the host is the same, we're in business. If not, there is
961 still hope -- read below. */
962 if (0 != strcasecmp (host, pconn.host))
964 /* Check if pconn.socket is talking to HOST under another name.
965 This happens often when both sites are virtual hosts
966 distinguished only by name and served by the same network
967 interface, and hence the same web server (possibly set up by
968 the ISP and serving many different web sites). This
969 admittedly unconventional optimization does not contradict
970 HTTP and works well with popular server software. */
974 struct address_list *al;
977 /* Don't try to talk to two different SSL sites over the same
978 secure connection! (Besides, it's not clear that
979 name-based virtual hosting is even possible with SSL.) */
982 /* If pconn.socket's peer is one of the IP addresses HOST
983 resolves to, pconn.socket is for all intents and purposes
984 already talking to HOST. */
986 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
988 /* Can't get the peer's address -- something must be very
989 wrong with the connection. */
990 invalidate_persistent ();
993 al = lookup_host (host, 0);
996 *host_lookup_failed = true;
1000 found = address_list_contains (al, &ip);
1001 address_list_release (al);
1006 /* The persistent connection's peer address was found among the
1007 addresses HOST resolved to; therefore, pconn.sock is in fact
1008 already talking to HOST -- no need to reconnect. */
1011 /* Finally, check whether the connection is still open. This is
1012 important because most servers implement liberal (short) timeout
1013 on persistent connections. Wget can of course always reconnect
1014 if the connection doesn't work out, but it's nicer to know in
1015 advance. This test is a logical followup of the first test, but
1016 is "expensive" and therefore placed at the end of the list.
1018 (Current implementation of test_socket_open has a nice side
1019 effect that it treats sockets with pending data as "closed".
1020 This is exactly what we want: if a broken server sends message
1021 body in response to HEAD, or if it sends more than conent-length
1022 data, we won't reuse the corrupted connection.) */
1024 if (!test_socket_open (pconn.socket))
1026 /* Oops, the socket is no longer open. Now that we know that,
1027 let's invalidate the persistent connection before returning
1029 invalidate_persistent ();
1036 /* The idea behind these two CLOSE macros is to distinguish between
1037 two cases: one when the job we've been doing is finished, and we
1038 want to close the connection and leave, and two when something is
1039 seriously wrong and we're closing the connection as part of
1042 In case of keep_alive, CLOSE_FINISH should leave the connection
1043 open, while CLOSE_INVALIDATE should still close it.
1045 Note that the semantics of the flag `keep_alive' is "this
1046 connection *will* be reused (the server has promised not to close
1047 the connection once we're done)", while the semantics of
1048 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1049 active, registered connection". */
1051 #define CLOSE_FINISH(fd) do { \
1054 if (pconn_active && (fd) == pconn.socket) \
1055 invalidate_persistent (); \
1064 #define CLOSE_INVALIDATE(fd) do { \
1065 if (pconn_active && (fd) == pconn.socket) \
1066 invalidate_persistent (); \
1074 wgint len; /* received length */
1075 wgint contlen; /* expected length */
1076 wgint restval; /* the restart value */
1077 int res; /* the result of last read */
1078 char *rderrmsg; /* error message from read error */
1079 char *newloc; /* new location (redirection) */
1080 char *remote_time; /* remote time-stamp string */
1081 char *error; /* textual HTTP error */
1082 int statcode; /* status code */
1083 wgint rd_size; /* amount of data read from socket */
1084 double dltime; /* time it took to download the data */
1085 const char *referer; /* value of the referer header. */
1086 char *local_file; /* local file name. */
1087 bool timestamp_checked; /* true if pre-download time-stamping checks
1088 * have already been performed */
1089 char *orig_file_name; /* name of file to compare for time-stamping
1090 * (might be != local_file if -K is set) */
1091 wgint orig_file_size; /* size of file to compare for time-stamping */
1092 time_t orig_file_tstamp; /* time-stamp of file to compare for
1097 free_hstat (struct http_stat *hs)
1099 xfree_null (hs->newloc);
1100 xfree_null (hs->remote_time);
1101 xfree_null (hs->error);
1102 xfree_null (hs->rderrmsg);
1103 xfree_null (hs->local_file);
1104 xfree_null (hs->orig_file_name);
1106 /* Guard against being called twice. */
1108 hs->remote_time = NULL;
1112 static char *create_authorization_line (const char *, const char *,
1113 const char *, const char *,
1114 const char *, bool *);
1115 static char *basic_authentication_encode (const char *, const char *);
1116 static bool known_authentication_scheme_p (const char *, const char *);
1117 static void load_cookies (void);
1119 #define BEGINS_WITH(line, string_constant) \
1120 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1121 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1122 || !line[sizeof (string_constant) - 1]))
1124 #define SET_USER_AGENT(req) do { \
1125 if (!opt.useragent) \
1126 request_set_header (req, "User-Agent", \
1127 aprintf ("Wget/%s", version_string), rel_value); \
1128 else if (*opt.useragent) \
1129 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1132 /* The flags that allow clobbering the file (opening with "wb").
1133 Defined here to avoid repetition later. #### This will require
1135 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1136 || opt.dirstruct || opt.output_document)
1138 /* Retrieve a document through HTTP protocol. It recognizes status
1139 code, and correctly handles redirections. It closes the network
1140 socket. If it receives an error from the functions below it, it
1141 will print it if there is enough information to do so (almost
1142 always), returning the error to the caller (i.e. http_loop).
1144 Various HTTP parameters are stored to hs.
1146 If PROXY is non-NULL, the connection will be made to the proxy
1147 server, and u->url will be requested. */
1149 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1151 struct request *req;
1154 char *user, *passwd;
1158 wgint contlen, contrange;
1165 /* Set to 1 when the authorization has failed permanently and should
1166 not be tried again. */
1167 bool auth_finished = false;
1169 /* Whether NTLM authentication is used for this request. */
1170 bool ntlm_seen = false;
1172 /* Whether our connection to the remote host is through SSL. */
1173 bool using_ssl = false;
1175 /* Whether a HEAD request will be issued (as opposed to GET or
1177 bool head_only = !!(*dt & HEAD_ONLY);
1180 struct response *resp;
1184 /* Whether this connection will be kept alive after the HTTP request
1188 /* Whether keep-alive should be inhibited.
1190 RFC 2068 requests that 1.0 clients not send keep-alive requests
1191 to proxies. This is because many 1.0 proxies do not interpret
1192 the Connection header and transfer it to the remote server,
1193 causing it to not close the connection and leave both the proxy
1194 and the client hanging. */
1195 bool inhibit_keep_alive =
1196 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1198 /* Headers sent when using POST. */
1199 wgint post_data_size = 0;
1201 bool host_lookup_failed = false;
1203 DEBUGP(("in gethttp 1\n"));
1206 if (u->scheme == SCHEME_HTTPS)
1208 /* Initialize the SSL context. After this has once been done,
1209 it becomes a no-op. */
1212 scheme_disable (SCHEME_HTTPS);
1213 logprintf (LOG_NOTQUIET,
1214 _("Disabling SSL due to encountered errors.\n"));
1215 return SSLINITFAILED;
1218 #endif /* HAVE_SSL */
1220 DEBUGP(("in gethttp 2\n"));
1221 DEBUGP(("in gethttp 3\n"));
1223 /* Initialize certain elements of struct http_stat. */
1227 hs->rderrmsg = NULL;
1229 hs->remote_time = NULL;
1234 /* Prepare the request to send. */
1236 req = request_new ();
1239 const char *meth = "GET";
1242 else if (opt.post_file_name || opt.post_data)
1244 /* Use the full path, i.e. one that includes the leading slash and
1245 the query string. E.g. if u->path is "foo/bar" and u->query is
1246 "param=value", full_path will be "/foo/bar?param=value". */
1249 /* When using SSL over proxy, CONNECT establishes a direct
1250 connection to the HTTPS server. Therefore use the same
1251 argument as when talking to the server directly. */
1252 && u->scheme != SCHEME_HTTPS
1255 meth_arg = xstrdup (u->url);
1257 meth_arg = url_full_path (u);
1258 request_set_method (req, meth, meth_arg);
1261 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1262 if (*dt & SEND_NOCACHE)
1263 request_set_header (req, "Pragma", "no-cache", rel_none);
1265 request_set_header (req, "Range",
1266 aprintf ("bytes=%s-",
1267 number_to_static_string (hs->restval)),
1269 SET_USER_AGENT (req);
1270 request_set_header (req, "Accept", "*/*", rel_none);
1272 /* Find the username and password for authentication. */
1275 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1276 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1277 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1281 /* We have the username and the password, but haven't tried
1282 any authorization yet. Let's see if the "Basic" method
1283 works. If not, we'll come back here and construct a
1284 proper authorization method with the right challenges.
1286 If we didn't employ this kind of logic, every URL that
1287 requires authorization would have to be processed twice,
1288 which is very suboptimal and generates a bunch of false
1289 "unauthorized" errors in the server log.
1291 #### But this logic also has a serious problem when used
1292 with stronger authentications: we *first* transmit the
1293 username and the password in clear text, and *then* attempt a
1294 stronger authentication scheme. That cannot be right! We
1295 are only fortunate that almost everyone still uses the
1296 `Basic' scheme anyway.
1298 There should be an option to prevent this from happening, for
1299 those who use strong authentication schemes and value their
1301 request_set_header (req, "Authorization",
1302 basic_authentication_encode (user, passwd),
1309 char *proxy_user, *proxy_passwd;
1310 /* For normal username and password, URL components override
1311 command-line/wgetrc parameters. With proxy
1312 authentication, it's the reverse, because proxy URLs are
1313 normally the "permanent" ones, so command-line args
1314 should take precedence. */
1315 if (opt.proxy_user && opt.proxy_passwd)
1317 proxy_user = opt.proxy_user;
1318 proxy_passwd = opt.proxy_passwd;
1322 proxy_user = proxy->user;
1323 proxy_passwd = proxy->passwd;
1325 /* #### This does not appear right. Can't the proxy request,
1326 say, `Digest' authentication? */
1327 if (proxy_user && proxy_passwd)
1328 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1330 /* If we're using a proxy, we will be connecting to the proxy
1334 /* Proxy authorization over SSL is handled below. */
1336 if (u->scheme != SCHEME_HTTPS)
1338 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1341 /* Generate the Host header, HOST:PORT. Take into account that:
1343 - Broken server-side software often doesn't recognize the PORT
1344 argument, so we must generate "Host: www.server.com" instead of
1345 "Host: www.server.com:80" (and likewise for https port).
1347 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1348 becomes ambiguous and needs to be rewritten as "Host:
1349 [3ffe:8100:200:2::2]:1234". */
1351 /* Formats arranged for hfmt[add_port][add_squares]. */
1352 static const char *hfmt[][2] = {
1353 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1355 int add_port = u->port != scheme_default_port (u->scheme);
1356 int add_squares = strchr (u->host, ':') != NULL;
1357 request_set_header (req, "Host",
1358 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1362 if (!inhibit_keep_alive)
1363 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1366 request_set_header (req, "Cookie",
1367 cookie_header (wget_cookie_jar,
1368 u->host, u->port, u->path,
1370 u->scheme == SCHEME_HTTPS
1377 if (opt.post_data || opt.post_file_name)
1379 request_set_header (req, "Content-Type",
1380 "application/x-www-form-urlencoded", rel_none);
1382 post_data_size = strlen (opt.post_data);
1385 post_data_size = file_size (opt.post_file_name);
1386 if (post_data_size == -1)
1388 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1389 opt.post_file_name, strerror (errno));
1393 request_set_header (req, "Content-Length",
1394 xstrdup (number_to_static_string (post_data_size)),
1398 /* Add the user headers. */
1399 if (opt.user_headers)
1402 for (i = 0; opt.user_headers[i]; i++)
1403 request_set_user_header (req, opt.user_headers[i]);
1407 /* We need to come back here when the initial attempt to retrieve
1408 without authorization header fails. (Expected to happen at least
1409 for the Digest authorization scheme.) */
1413 /* Establish the connection. */
1415 if (!inhibit_keep_alive)
1417 /* Look for a persistent connection to target host, unless a
1418 proxy is used. The exception is when SSL is in use, in which
1419 case the proxy is nothing but a passthrough to the target
1420 host, registered as a connection to the latter. */
1421 struct url *relevant = conn;
1423 if (u->scheme == SCHEME_HTTPS)
1427 if (persistent_available_p (relevant->host, relevant->port,
1429 relevant->scheme == SCHEME_HTTPS,
1433 &host_lookup_failed))
1435 sock = pconn.socket;
1436 using_ssl = pconn.ssl;
1437 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1438 escnonprint (pconn.host), pconn.port);
1439 DEBUGP (("Reusing fd %d.\n", sock));
1440 if (pconn.authorized)
1441 /* If the connection is already authorized, the "Basic"
1442 authorization added by code above is unnecessary and
1444 request_remove_header (req, "Authorization");
1450 /* In its current implementation, persistent_available_p will
1451 look up conn->host in some cases. If that lookup failed, we
1452 don't need to bother with connect_to_host. */
1453 if (host_lookup_failed)
1459 sock = connect_to_host (conn->host, conn->port);
1468 return (retryable_socket_connect_error (errno)
1469 ? CONERROR : CONIMPOSSIBLE);
1473 if (proxy && u->scheme == SCHEME_HTTPS)
1475 /* When requesting SSL URLs through proxies, use the
1476 CONNECT method to request passthrough. */
1477 struct request *connreq = request_new ();
1478 request_set_method (connreq, "CONNECT",
1479 aprintf ("%s:%d", u->host, u->port));
1480 SET_USER_AGENT (connreq);
1483 request_set_header (connreq, "Proxy-Authorization",
1484 proxyauth, rel_value);
1485 /* Now that PROXYAUTH is part of the CONNECT request,
1486 zero it out so we don't send proxy authorization with
1487 the regular request below. */
1490 /* Examples in rfc2817 use the Host header in CONNECT
1491 requests. I don't see how that gains anything, given
1492 that the contents of Host would be exactly the same as
1493 the contents of CONNECT. */
1495 write_error = request_send (connreq, sock);
1496 request_free (connreq);
1497 if (write_error < 0)
1499 CLOSE_INVALIDATE (sock);
1503 head = read_http_response_head (sock);
1506 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1508 CLOSE_INVALIDATE (sock);
1517 DEBUGP (("proxy responded with: [%s]\n", head));
1519 resp = resp_new (head);
1520 statcode = resp_status (resp, &message);
1523 if (statcode != 200)
1526 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1527 message ? escnonprint (message) : "?");
1528 xfree_null (message);
1531 xfree_null (message);
1533 /* SOCK is now *really* connected to u->host, so update CONN
1534 to reflect this. That way register_persistent will
1535 register SOCK as being connected to u->host:u->port. */
1539 if (conn->scheme == SCHEME_HTTPS)
1541 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1548 #endif /* HAVE_SSL */
1551 /* Send the request to server. */
1552 write_error = request_send (req, sock);
1554 if (write_error >= 0)
1558 DEBUGP (("[POST data: %s]\n", opt.post_data));
1559 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1561 else if (opt.post_file_name && post_data_size != 0)
1562 write_error = post_file (sock, opt.post_file_name, post_data_size);
1565 if (write_error < 0)
1567 CLOSE_INVALIDATE (sock);
1571 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1572 proxy ? "Proxy" : "HTTP");
1577 head = read_http_response_head (sock);
1582 logputs (LOG_NOTQUIET, _("No data received.\n"));
1583 CLOSE_INVALIDATE (sock);
1589 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1591 CLOSE_INVALIDATE (sock);
1596 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1598 resp = resp_new (head);
1600 /* Check for status line. */
1602 statcode = resp_status (resp, &message);
1603 if (!opt.server_response)
1604 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1605 message ? escnonprint (message) : "");
1608 logprintf (LOG_VERBOSE, "\n");
1609 print_server_response (resp, " ");
1612 DEBUGP(("in gethttp 4\n"));
1614 /* Determine the local filename if needed. Notice that if -O is used
1615 * hstat.local_file is set by http_loop to the argument of -O. */
1616 if (!hs->local_file)
1618 if (resp_header_copy (resp, "Content-Disposition", hdrval, sizeof (hdrval)))
1619 /* Honor Content-Disposition. */
1621 hs->local_file = xstrdup (hdrval);
1624 /* Choose filename according to URL name. */
1626 hs->local_file = url_file_name (u);
1630 DEBUGP(("in gethttp 5\n"));
1632 /* TODO: perform this check only once. */
1633 if (opt.noclobber && file_exists_p (hs->local_file))
1635 /* If opt.noclobber is turned on and file already exists, do not
1636 retrieve the file */
1637 logprintf (LOG_VERBOSE, _("\
1638 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1639 /* If the file is there, we suppose it's retrieved OK. */
1642 /* #### Bogusness alert. */
1643 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1644 if (has_html_suffix_p (hs->local_file))
1650 /* Support timestamping */
1651 /* TODO: move this code out of gethttp. */
1652 if (opt.timestamping && !hs->timestamp_checked)
1654 size_t filename_len = strlen (hs->local_file);
1655 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1656 bool local_dot_orig_file_exists = false;
1657 char *local_filename = NULL;
1660 if (opt.backup_converted)
1661 /* If -K is specified, we'll act on the assumption that it was specified
1662 last time these files were downloaded as well, and instead of just
1663 comparing local file X against server file X, we'll compare local
1664 file X.orig (if extant, else X) against server file X. If -K
1665 _wasn't_ specified last time, or the server contains files called
1666 *.orig, -N will be back to not operating correctly with -k. */
1668 /* Would a single s[n]printf() call be faster? --dan
1670 Definitely not. sprintf() is horribly slow. It's a
1671 different question whether the difference between the two
1672 affects a program. Usually I'd say "no", but at one
1673 point I profiled Wget, and found that a measurable and
1674 non-negligible amount of time was lost calling sprintf()
1675 in url.c. Replacing sprintf with inline calls to
1676 strcpy() and number_to_string() made a difference.
1678 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1679 memcpy (filename_plus_orig_suffix + filename_len,
1680 ".orig", sizeof (".orig"));
1682 /* Try to stat() the .orig file. */
1683 if (stat (filename_plus_orig_suffix, &st) == 0)
1685 local_dot_orig_file_exists = 1;
1686 local_filename = filename_plus_orig_suffix;
1690 if (!local_dot_orig_file_exists)
1691 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1692 if (stat (hs->local_file, &st) == 0)
1693 local_filename = hs->local_file;
1695 if (local_filename != NULL)
1696 /* There was a local file, so we'll check later to see if the version
1697 the server has is the same version we already have, allowing us to
1700 hs->orig_file_name = xstrdup (local_filename);
1701 hs->orig_file_size = st.st_size;
1702 hs->orig_file_tstamp = st.st_mtime;
1704 /* Modification time granularity is 2 seconds for Windows, so
1705 increase local time by 1 second for later comparison. */
1706 ++hs->orig_file_tstamp;
1711 if (!opt.ignore_length
1712 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1716 parsed = str_to_wgint (hdrval, NULL, 10);
1717 if (parsed == WGINT_MAX && errno == ERANGE)
1719 #### If Content-Length is out of range, it most likely
1720 means that the file is larger than 2G and that we're
1721 compiled without LFS. In that case we should probably
1722 refuse to even attempt to download the file. */
1728 /* Check for keep-alive related responses. */
1729 if (!inhibit_keep_alive && contlen != -1)
1731 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1733 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1735 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1740 /* The server has promised that it will not close the connection
1741 when we're done. This means that we can register it. */
1742 register_persistent (conn->host, conn->port, sock, using_ssl);
1744 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1746 /* Authorization is required. */
1747 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1748 CLOSE_FINISH (sock);
1750 CLOSE_INVALIDATE (sock);
1751 pconn.authorized = false;
1752 if (!auth_finished && (user && passwd))
1754 /* IIS sends multiple copies of WWW-Authenticate, one with
1755 the value "negotiate", and other(s) with data. Loop over
1756 all the occurrences and pick the one we recognize. */
1758 const char *wabeg, *waend;
1759 char *www_authenticate = NULL;
1761 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1762 &wabeg, &waend)) != -1;
1764 if (known_authentication_scheme_p (wabeg, waend))
1766 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1770 if (!www_authenticate)
1771 /* If the authentication header is missing or
1772 unrecognized, there's no sense in retrying. */
1773 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1774 else if (BEGINS_WITH (www_authenticate, "Basic"))
1775 /* If the authentication scheme is "Basic", which we send
1776 by default, there's no sense in retrying either. (This
1777 should be changed when we stop sending "Basic" data by
1783 pth = url_full_path (u);
1784 request_set_header (req, "Authorization",
1785 create_authorization_line (www_authenticate,
1787 request_method (req),
1791 if (BEGINS_WITH (www_authenticate, "NTLM"))
1794 goto retry_with_auth;
1797 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1801 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1803 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1805 pconn.authorized = true;
1809 hs->statcode = statcode;
1811 hs->error = xstrdup (_("Malformed status line"));
1813 hs->error = xstrdup (_("(no description)"));
1815 hs->error = xstrdup (message);
1816 xfree_null (message);
1818 type = resp_header_strdup (resp, "Content-Type");
1821 char *tmp = strchr (type, ';');
1824 while (tmp > type && ISSPACE (tmp[-1]))
1829 hs->newloc = resp_header_strdup (resp, "Location");
1830 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1832 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1836 const char *scbeg, *scend;
1837 /* The jar should have been created by now. */
1838 assert (wget_cookie_jar != NULL);
1840 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1841 &scbeg, &scend)) != -1;
1844 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1845 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1846 u->path, set_cookie);
1850 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1852 wgint first_byte_pos, last_byte_pos, entity_length;
1853 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1855 contrange = first_byte_pos;
1859 /* 20x responses are counted among successful by default. */
1860 if (H_20X (statcode))
1863 /* Return if redirected. */
1864 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1866 /* RFC2068 says that in case of the 300 (multiple choices)
1867 response, the server can output a preferred URL through
1868 `Location' header; otherwise, the request should be treated
1869 like GET. So, if the location is set, it will be a
1870 redirection; otherwise, just proceed normally. */
1871 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1875 logprintf (LOG_VERBOSE,
1876 _("Location: %s%s\n"),
1877 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1878 hs->newloc ? _(" [following]") : "");
1879 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1880 CLOSE_FINISH (sock);
1882 CLOSE_INVALIDATE (sock);
1888 /* If content-type is not given, assume text/html. This is because
1889 of the multitude of broken CGI's that "forget" to generate the
1892 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1893 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1898 if (opt.html_extension && (*dt & TEXTHTML))
1899 /* -E / --html-extension / html_extension = on was specified, and this is a
1900 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1901 already the file's suffix, tack on ".html". */
1903 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
1905 if (last_period_in_local_filename == NULL
1906 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1907 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1909 int local_filename_len = strlen (hs->local_file);
1910 /* Resize the local file, allowing for ".html" preceded by
1911 optional ".NUMBER". */
1912 hs->local_file = xrealloc (hs->local_file,
1913 local_filename_len + 24 + sizeof (".html"));
1914 strcpy(hs->local_file + local_filename_len, ".html");
1915 /* If clobbering is not allowed and the file, as named,
1916 exists, tack on ".NUMBER.html" instead. */
1921 sprintf (hs->local_file + local_filename_len,
1922 ".%d.html", ext_num++);
1923 while (file_exists_p (hs->local_file));
1925 *dt |= ADDED_HTML_EXTENSION;
1929 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1931 /* If `-c' is in use and the file has been fully downloaded (or
1932 the remote file has shrunk), Wget effectively requests bytes
1933 after the end of file and the server response with 416. */
1934 logputs (LOG_VERBOSE, _("\
1935 \n The file is already fully retrieved; nothing to do.\n\n"));
1936 /* In case the caller inspects. */
1939 /* Mark as successfully retrieved. */
1942 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1943 might be more bytes in the body. */
1944 return RETRUNNEEDED;
1946 if ((contrange != 0 && contrange != hs->restval)
1947 || (H_PARTIAL (statcode) && !contrange))
1949 /* The Range request was somehow misunderstood by the server.
1952 CLOSE_INVALIDATE (sock);
1955 hs->contlen = contlen + contrange;
1961 /* No need to print this output if the body won't be
1962 downloaded at all, or if the original server response is
1964 logputs (LOG_VERBOSE, _("Length: "));
1967 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
1968 if (contlen + contrange >= 1024)
1969 logprintf (LOG_VERBOSE, " (%s)",
1970 human_readable (contlen + contrange));
1973 if (contlen >= 1024)
1974 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1975 number_to_static_string (contlen),
1976 human_readable (contlen));
1978 logprintf (LOG_VERBOSE, _(", %s remaining"),
1979 number_to_static_string (contlen));
1983 logputs (LOG_VERBOSE,
1984 opt.ignore_length ? _("ignored") : _("unspecified"));
1986 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1988 logputs (LOG_VERBOSE, "\n");
1992 type = NULL; /* We don't need it any more. */
1994 /* Return if we have no intention of further downloading. */
1995 if (!(*dt & RETROKF) || head_only)
1997 /* In case the caller cares to look... */
2002 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2003 servers not to send body in response to a HEAD request, and
2004 those that do will likely be caught by test_socket_open.
2005 If not, they can be worked around using
2006 `--no-http-keep-alive'. */
2007 CLOSE_FINISH (sock);
2008 else if (keep_alive && skip_short_body (sock, contlen))
2009 /* Successfully skipped the body; also keep using the socket. */
2010 CLOSE_FINISH (sock);
2012 CLOSE_INVALIDATE (sock);
2013 return RETRFINISHED;
2016 /* Open the local file. */
2019 mkalldirs (hs->local_file);
2021 rotate_backups (hs->local_file);
2023 fp = fopen (hs->local_file, "ab");
2024 else if (ALLOW_CLOBBER)
2025 fp = fopen (hs->local_file, "wb");
2028 fp = fopen_excl (hs->local_file, true);
2029 if (!fp && errno == EEXIST)
2031 /* We cannot just invent a new name and use it (which is
2032 what functions like unique_create typically do)
2033 because we told the user we'd use this name.
2034 Instead, return and retry the download. */
2035 logprintf (LOG_NOTQUIET,
2036 _("%s has sprung into existence.\n"),
2038 CLOSE_INVALIDATE (sock);
2039 return FOPEN_EXCL_ERR;
2044 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2045 CLOSE_INVALIDATE (sock);
2052 /* This confuses the timestamping code that checks for file size.
2053 #### The timestamping code should be smarter about file size. */
2054 if (opt.save_headers && hs->restval == 0)
2055 fwrite (head, 1, strlen (head), fp);
2057 /* Now we no longer need to store the response header. */
2060 /* Download the request body. */
2063 /* If content-length is present, read that much; otherwise, read
2064 until EOF. The HTTP spec doesn't require the server to
2065 actually close the connection when it's done sending data. */
2066 flags |= rb_read_exactly;
2067 if (hs->restval > 0 && contrange == 0)
2068 /* If the server ignored our range request, instruct fd_read_body
2069 to skip the first RESTVAL bytes of body. */
2070 flags |= rb_skip_startpos;
2071 hs->len = hs->restval;
2073 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2074 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2078 CLOSE_FINISH (sock);
2082 hs->rderrmsg = xstrdup (fd_errstr (sock));
2083 CLOSE_INVALIDATE (sock);
2090 return RETRFINISHED;
2093 /* The genuine HTTP loop! This is the part where the retrieval is
2094 retried, and retried, and retried, and... */
2096 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2097 int *dt, struct url *proxy)
2100 bool got_head = false; /* used for time-stamping */
2104 time_t tmr = -1; /* remote time-stamp */
2105 wgint local_size = 0; /* the size of the local file */
2106 struct http_stat hstat; /* HTTP status */
2109 DEBUGP(("in http_loop\n"));
2111 /* Assert that no value for *LOCAL_FILE was passed. */
2112 assert (local_file == NULL || *local_file == NULL);
2114 /* Set LOCAL_FILE parameter. */
2115 if (local_file && opt.output_document)
2116 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2118 /* Reset NEWLOC parameter. */
2121 /* This used to be done in main(), but it's a better idea to do it
2122 here so that we don't go through the hoops if we're just using
2127 /* Warn on (likely bogus) wildcard usage in HTTP. */
2128 if (opt.ftp_glob && has_wildcards_p (u->path))
2129 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2131 /* Setup hstat struct. */
2133 hstat.referer = referer;
2135 if (opt.output_document)
2136 hstat.local_file = xstrdup (opt.output_document);
2138 /* Reset the counter. */
2141 /* Reset the document type. */
2147 DEBUGP(("in http_loop LOOP\n"));
2149 /* Increment the pass counter. */
2151 sleep_between_retrievals (count);
2153 /* Get the current time string. */
2154 tms = time_str (NULL);
2156 /* Print fetch message, if opt.verbose. */
2159 char *hurl = url_string (u, true);
2163 sprintf (tmp, _("(try:%2d)"), count);
2164 logprintf (LOG_VERBOSE, "--%s-- %s\n %s\n",
2167 ws_changetitle (hurl);
2172 /* Default document type is empty. However, if spider mode is
2173 on or time-stamping is employed, HEAD_ONLY commands is
2174 encoded within *dt. */
2175 if (opt.spider || (opt.timestamping && !got_head))
2180 /* Decide whether or not to restart. */
2182 && stat (hstat.local_file, &st) == 0
2183 && S_ISREG (st.st_mode))
2184 /* When -c is used, continue from on-disk size. (Can't use
2185 hstat.len even if count>1 because we don't want a failed
2186 first attempt to clobber existing data.) */
2187 hstat.restval = st.st_size;
2189 /* otherwise, continue where the previous try left off */
2190 hstat.restval = hstat.len;
2194 /* Decide whether to send the no-cache directive. We send it in
2196 a) we're using a proxy, and we're past our first retrieval.
2197 Some proxies are notorious for caching incomplete data, so
2198 we require a fresh get.
2199 b) caching is explicitly inhibited. */
2200 if ((proxy && count > 1) /* a */
2201 || !opt.allow_cache /* b */
2203 *dt |= SEND_NOCACHE;
2205 *dt &= ~SEND_NOCACHE;
2207 /* Try fetching the document, or at least its head. */
2208 err = gethttp (u, &hstat, dt, proxy);
2211 tms = time_str (NULL);
2213 /* Get the new location (with or without the redirection). */
2215 *newloc = xstrdup (hstat.newloc);
2219 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2220 case CONERROR: case READERR: case WRITEFAILED:
2221 case RANGEERR: case FOPEN_EXCL_ERR:
2222 /* Non-fatal errors continue executing the loop, which will
2223 bring them to "while" statement at the end, to judge
2224 whether the number of tries was exceeded. */
2225 /* free_hstat (&hstat); */
2226 printwhat (count, opt.ntry);
2228 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2229 case SSLINITFAILED: case CONTNOTSUPPORTED:
2230 /* Fatal errors just return from the function. */
2231 free_hstat (&hstat);
2233 case FWRITEERR: case FOPENERR:
2234 /* Another fatal error. */
2235 logputs (LOG_VERBOSE, "\n");
2236 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2237 hstat.local_file, strerror (errno));
2238 free_hstat (&hstat);
2241 /* Another fatal error. */
2242 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2243 free_hstat (&hstat);
2246 /* Return the new location to the caller. */
2249 logprintf (LOG_NOTQUIET,
2250 _("ERROR: Redirection (%d) without location.\n"),
2252 free_hstat (&hstat);
2255 free_hstat (&hstat);
2258 /* The file was already fully retrieved. */
2259 free_hstat (&hstat);
2262 /* Deal with you later. */
2265 /* All possibilities should have been exhausted. */
2269 if (!(*dt & RETROKF))
2273 /* #### Ugly ugly ugly! */
2274 char *hurl = url_string (u, true);
2275 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2278 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2279 tms, hstat.statcode, escnonprint (hstat.error));
2280 logputs (LOG_VERBOSE, "\n");
2281 free_hstat (&hstat);
2285 /* Did we get the time-stamp? */
2288 if (opt.timestamping && !hstat.remote_time)
2290 logputs (LOG_NOTQUIET, _("\
2291 Last-modified header missing -- time-stamps turned off.\n"));
2293 else if (hstat.remote_time)
2295 /* Convert the date-string into struct tm. */
2296 tmr = http_atotm (hstat.remote_time);
2297 if (tmr == (time_t) (-1))
2298 logputs (LOG_VERBOSE, _("\
2299 Last-modified header invalid -- time-stamp ignored.\n"));
2303 /* The time-stamping section. */
2304 if (opt.timestamping && !got_head)
2306 got_head = true; /* no more time-stamping */
2308 count = 0; /* the retrieve count for HEAD is reset */
2310 if (hstat.remote_time && tmr != (time_t) (-1))
2312 /* Now time-stamping can be used validly. Time-stamping
2313 means that if the sizes of the local and remote file
2314 match, and local file is newer than the remote file,
2315 it will not be retrieved. Otherwise, the normal
2316 download procedure is resumed. */
2317 if (hstat.orig_file_tstamp >= tmr)
2319 if (hstat.contlen == -1 || hstat.orig_file_size == hstat.contlen)
2321 logprintf (LOG_VERBOSE, _("\
2322 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2323 hstat.orig_file_name);
2324 free_hstat (&hstat);
2329 logprintf (LOG_VERBOSE, _("\
2330 The sizes do not match (local %s) -- retrieving.\n"),
2331 number_to_static_string (local_size));
2335 logputs (LOG_VERBOSE,
2336 _("Remote file is newer, retrieving.\n"));
2339 /* free_hstat (&hstat); */
2340 hstat.timestamp_checked = true;
2344 if ((tmr != (time_t) (-1))
2346 && ((hstat.len == hstat.contlen) ||
2347 ((hstat.res == 0) && (hstat.contlen == -1))))
2349 /* #### This code repeats in http.c and ftp.c. Move it to a
2351 const char *fl = NULL;
2352 if (opt.output_document)
2354 if (output_stream_regular)
2355 fl = opt.output_document;
2358 fl = hstat.local_file;
2362 /* End of time-stamping section. */
2366 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2367 escnonprint (hstat.error));
2371 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2372 total_download_time += hstat.dltime;
2374 if (hstat.len == hstat.contlen)
2378 logprintf (LOG_VERBOSE,
2379 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2380 tms, tmrate, hstat.local_file,
2381 number_to_static_string (hstat.len),
2382 number_to_static_string (hstat.contlen));
2383 logprintf (LOG_NONVERBOSE,
2384 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2386 number_to_static_string (hstat.len),
2387 number_to_static_string (hstat.contlen),
2388 hstat.local_file, count);
2391 total_downloaded_bytes += hstat.len;
2393 /* Remember that we downloaded the file for later ".orig" code. */
2394 if (*dt & ADDED_HTML_EXTENSION)
2395 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2397 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2399 free_hstat (&hstat);
2402 else if (hstat.res == 0) /* No read error */
2404 if (hstat.contlen == -1) /* We don't know how much we were supposed
2405 to get, so assume we succeeded. */
2409 logprintf (LOG_VERBOSE,
2410 _("%s (%s) - `%s' saved [%s]\n\n"),
2411 tms, tmrate, hstat.local_file,
2412 number_to_static_string (hstat.len));
2413 logprintf (LOG_NONVERBOSE,
2414 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2415 tms, u->url, number_to_static_string (hstat.len),
2416 hstat.local_file, count);
2419 total_downloaded_bytes += hstat.len;
2421 /* Remember that we downloaded the file for later ".orig" code. */
2422 if (*dt & ADDED_HTML_EXTENSION)
2423 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2425 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2427 free_hstat (&hstat);
2430 else if (hstat.len < hstat.contlen) /* meaning we lost the
2431 connection too soon */
2433 logprintf (LOG_VERBOSE,
2434 _("%s (%s) - Connection closed at byte %s. "),
2435 tms, tmrate, number_to_static_string (hstat.len));
2436 printwhat (count, opt.ntry);
2437 /* free_hstat (&hstat); */
2441 /* Getting here would mean reading more data than
2442 requested with content-length, which we never do. */
2445 else /* from now on hstat.res can only be -1 */
2447 if (hstat.contlen == -1)
2449 logprintf (LOG_VERBOSE,
2450 _("%s (%s) - Read error at byte %s (%s)."),
2451 tms, tmrate, number_to_static_string (hstat.len),
2453 printwhat (count, opt.ntry);
2454 /* free_hstat (&hstat); */
2457 else /* hstat.res == -1 and contlen is given */
2459 logprintf (LOG_VERBOSE,
2460 _("%s (%s) - Read error at byte %s/%s (%s). "),
2462 number_to_static_string (hstat.len),
2463 number_to_static_string (hstat.contlen),
2465 printwhat (count, opt.ntry);
2466 /* free_hstat (&hstat); */
2472 while (!opt.ntry || (count < opt.ntry));
2477 /* Check whether the result of strptime() indicates success.
2478 strptime() returns the pointer to how far it got to in the string.
2479 The processing has been successful if the string is at `GMT' or
2480 `+X', or at the end of the string.
2482 In extended regexp parlance, the function returns 1 if P matches
2483 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2484 can return) is considered a failure and 0 is returned. */
2486 check_end (const char *p)
2490 while (ISSPACE (*p))
2493 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2494 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2500 /* Convert the textual specification of time in TIME_STRING to the
2501 number of seconds since the Epoch.
2503 TIME_STRING can be in any of the three formats RFC2616 allows the
2504 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2505 as well as the time format used in the Set-Cookie header.
2506 Timezones are ignored, and should be GMT.
2508 Return the computed time_t representation, or -1 if the conversion
2511 This function uses strptime with various string formats for parsing
2512 TIME_STRING. This results in a parser that is not as lenient in
2513 interpreting TIME_STRING as I would like it to be. Being based on
2514 strptime, it always allows shortened months, one-digit days, etc.,
2515 but due to the multitude of formats in which time can be
2516 represented, an ideal HTTP time parser would be even more
2517 forgiving. It should completely ignore things like week days and
2518 concentrate only on the various forms of representing years,
2519 months, days, hours, minutes, and seconds. For example, it would
2520 be nice if it accepted ISO 8601 out of the box.
2522 I've investigated free and PD code for this purpose, but none was
2523 usable. getdate was big and unwieldy, and had potential copyright
2524 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2525 distributed with phttpd, is excellent, but we cannot use it because
2526 it is not assigned to the FSF. So I stuck it with strptime. */
2529 http_atotm (const char *time_string)
2531 /* NOTE: Solaris strptime man page claims that %n and %t match white
2532 space, but that's not universally available. Instead, we simply
2533 use ` ' to mean "skip all WS", which works under all strptime
2534 implementations I've tested. */
2536 static const char *time_formats[] = {
2537 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2538 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2539 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2540 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2541 (used in Set-Cookie, defined in the
2542 Netscape cookie specification.) */
2544 const char *oldlocale;
2546 time_t ret = (time_t) -1;
2548 /* Solaris strptime fails to recognize English month names in
2549 non-English locales, which we work around by temporarily setting
2550 locale to C before invoking strptime. */
2551 oldlocale = setlocale (LC_TIME, NULL);
2552 setlocale (LC_TIME, "C");
2554 for (i = 0; i < countof (time_formats); i++)
2558 /* Some versions of strptime use the existing contents of struct
2559 tm to recalculate the date according to format. Zero it out
2560 to prevent stack garbage from influencing strptime. */
2563 if (check_end (strptime (time_string, time_formats[i], &t)))
2570 /* Restore the previous locale. */
2571 setlocale (LC_TIME, oldlocale);
2576 /* Authorization support: We support three authorization schemes:
2578 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2580 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2581 consisting of answering to the server's challenge with the proper
2584 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2585 Stenberg for libcurl. Like digest, NTLM is based on a
2586 challenge-response mechanism, but unlike digest, it is non-standard
2587 (authenticates TCP connections rather than requests), undocumented
2588 and Microsoft-specific. */
2590 /* Create the authentication header contents for the `Basic' scheme.
2591 This is done by encoding the string "USER:PASS" to base64 and
2592 prepending the string "Basic " in front of it. */
2595 basic_authentication_encode (const char *user, const char *passwd)
2598 int len1 = strlen (user) + 1 + strlen (passwd);
2600 t1 = (char *)alloca (len1 + 1);
2601 sprintf (t1, "%s:%s", user, passwd);
2603 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2604 base64_encode (t1, len1, t2);
2606 return concat_strings ("Basic ", t2, (char *) 0);
2609 #define SKIP_WS(x) do { \
2610 while (ISSPACE (*(x))) \
2614 #ifdef ENABLE_DIGEST
2615 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2616 of a field in such a header. If the field is the one specified by
2617 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2618 digest authorization code), extract its value in the (char*)
2619 variable pointed by RET. Returns negative on a malformed header,
2620 or number of bytes that have been parsed by this call. */
2622 extract_header_attr (const char *au, const char *attr_name, char **ret)
2625 const char *cp = au;
2627 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2629 cp += strlen (attr_name);
2642 for (ep = cp; *ep && *ep != '\"'; ep++)
2647 *ret = strdupdelim (cp, ep);
2654 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2655 an array of 16 bytes containing the hash keys, and BUF should be a
2656 buffer of 33 writable characters (32 for hex digits plus one for
2657 zero termination). */
2659 dump_hash (char *buf, const unsigned char *hash)
2663 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2665 *buf++ = XNUM_TO_digit (*hash >> 4);
2666 *buf++ = XNUM_TO_digit (*hash & 0xf);
2671 /* Take the line apart to find the challenge, and compose a digest
2672 authorization header. See RFC2069 section 2.1.2. */
2674 digest_authentication_encode (const char *au, const char *user,
2675 const char *passwd, const char *method,
2678 static char *realm, *opaque, *nonce;
2683 { "realm", &realm },
2684 { "opaque", &opaque },
2689 realm = opaque = nonce = NULL;
2691 au += 6; /* skip over `Digest' */
2697 for (i = 0; i < countof (options); i++)
2699 int skip = extract_header_attr (au, options[i].name,
2700 options[i].variable);
2704 xfree_null (opaque);
2714 if (i == countof (options))
2716 while (*au && *au != '=')
2724 while (*au && *au != '\"')
2731 while (*au && *au != ',')
2736 if (!realm || !nonce || !user || !passwd || !path || !method)
2739 xfree_null (opaque);
2744 /* Calculate the digest value. */
2746 ALLOCA_MD5_CONTEXT (ctx);
2747 unsigned char hash[MD5_HASHLEN];
2748 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2749 char response_digest[MD5_HASHLEN * 2 + 1];
2751 /* A1BUF = H(user ":" realm ":" password) */
2753 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2754 gen_md5_update ((unsigned char *)":", 1, ctx);
2755 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2756 gen_md5_update ((unsigned char *)":", 1, ctx);
2757 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2758 gen_md5_finish (ctx, hash);
2759 dump_hash (a1buf, hash);
2761 /* A2BUF = H(method ":" path) */
2763 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2764 gen_md5_update ((unsigned char *)":", 1, ctx);
2765 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2766 gen_md5_finish (ctx, hash);
2767 dump_hash (a2buf, hash);
2769 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2771 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2772 gen_md5_update ((unsigned char *)":", 1, ctx);
2773 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2774 gen_md5_update ((unsigned char *)":", 1, ctx);
2775 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2776 gen_md5_finish (ctx, hash);
2777 dump_hash (response_digest, hash);
2779 res = xmalloc (strlen (user)
2784 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2785 + (opaque ? strlen (opaque) : 0)
2787 sprintf (res, "Digest \
2788 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2789 user, realm, nonce, path, response_digest);
2792 char *p = res + strlen (res);
2793 strcat (p, ", opaque=\"");
2800 #endif /* ENABLE_DIGEST */
2802 /* Computing the size of a string literal must take into account that
2803 value returned by sizeof includes the terminating \0. */
2804 #define STRSIZE(literal) (sizeof (literal) - 1)
2806 /* Whether chars in [b, e) begin with the literal string provided as
2807 first argument and are followed by whitespace or terminating \0.
2808 The comparison is case-insensitive. */
2809 #define STARTS(literal, b, e) \
2810 ((e) - (b) >= STRSIZE (literal) \
2811 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2812 && ((e) - (b) == STRSIZE (literal) \
2813 || ISSPACE (b[STRSIZE (literal)])))
2816 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2818 return STARTS ("Basic", hdrbeg, hdrend)
2819 #ifdef ENABLE_DIGEST
2820 || STARTS ("Digest", hdrbeg, hdrend)
2823 || STARTS ("NTLM", hdrbeg, hdrend)
2830 /* Create the HTTP authorization request header. When the
2831 `WWW-Authenticate' response header is seen, according to the
2832 authorization scheme specified in that header (`Basic' and `Digest'
2833 are supported by the current implementation), produce an
2834 appropriate HTTP authorization request header. */
2836 create_authorization_line (const char *au, const char *user,
2837 const char *passwd, const char *method,
2838 const char *path, bool *finished)
2840 /* We are called only with known schemes, so we can dispatch on the
2842 switch (TOUPPER (*au))
2844 case 'B': /* Basic */
2846 return basic_authentication_encode (user, passwd);
2847 #ifdef ENABLE_DIGEST
2848 case 'D': /* Digest */
2850 return digest_authentication_encode (au, user, passwd, method, path);
2853 case 'N': /* NTLM */
2854 if (!ntlm_input (&pconn.ntlm, au))
2859 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2862 /* We shouldn't get here -- this function should be only called
2863 with values approved by known_authentication_scheme_p. */
2871 if (!wget_cookie_jar)
2872 wget_cookie_jar = cookie_jar_new ();
2873 if (opt.cookies_input && !cookies_loaded_p)
2875 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2876 cookies_loaded_p = true;
2883 if (wget_cookie_jar)
2884 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2890 xfree_null (pconn.host);
2891 if (wget_cookie_jar)
2892 cookie_jar_delete (wget_cookie_jar);